runtime/analyze-conversation: route through callSite: 'analyzeConversation'

assistant/src/runtime/services/__tests__/analyze-conversation.test.ts

@@ -50,23 +50,6 @@ mock.module("../../../export/transcript-formatter.js", () => ({
   buildAnalysisTranscript: () => "user: hi",
 }));
 
-// Default config stub — individual tests can override via mockGetConfig.
-interface AnalysisConfigStub {
-  analysis: {
-    modelIntent?: string;
-    modelOverride?: string;
-  };
-}
-const mockGetConfig = mock(
-  (): AnalysisConfigStub => ({
-    analysis: {},
-  }),
-);
-
-mock.module("../../../config/loader.js", () => ({
-  getConfig: mockGetConfig,
-}));
-
 import { AssistantEventHub } from "../../assistant-event-hub.js";
 import type { SendMessageDeps } from "../../http-types.js";
 import { analyzeConversation } from "../analyze-conversation.js";
@@ -90,8 +73,6 @@ beforeEach(() => {
   mockFindAnalysisConversationFor.mockImplementation(() => null);
   mockGetConversationSource.mockReset();
   mockGetConversationSource.mockImplementation(() => null);
-  mockGetConfig.mockReset();
-  mockGetConfig.mockImplementation(() => ({ analysis: {} }));
 });
 
 function makeConversation() {
@@ -221,12 +202,17 @@ describe("analyzeConversation", () => {
     expect(allowedTools).toBeInstanceOf(Set);
     expect(allowedTools?.size).toBe(0);
 
-    // Fires the agent loop.
+    // Fires the agent loop with the analyzeConversation call-site so the
+    // per-call provider config flows through `resolveCallSiteConfig`.
     expect(conversation.runAgentLoop).toHaveBeenCalledWith(
       expect.any(String),
       "msg-1",
       expect.any(Function),
-      expect.objectContaining({ isInteractive: false, isUserMessage: true }),
+      expect.objectContaining({
+        isInteractive: false,
+        isUserMessage: true,
+        callSite: "analyzeConversation",
+      }),
     );
   });
 
@@ -405,40 +391,36 @@ describe("analyzeConversation", () => {
     expect(mockAddMessage).not.toHaveBeenCalled();
   });
 
-  test("auto: passes modelOverride through to getOrCreateConversation when set in config", async () => {
-    mockGetConfig.mockImplementation(() => ({
-      analysis: {
-        modelIntent: "quality-optimized",
-        modelOverride: "claude-opus-4-6",
-      },
-    }));
+  test("auto: routes the agent loop through callSite: 'analyzeConversation'", async () => {
     const conversation = makeConversation();
     const deps = makeDeps(conversation);
 
     await analyzeConversation("conv-1", deps, { trigger: "auto" });
 
-    expect(deps.getOrCreateConversation).toHaveBeenCalledWith(
-      "analysis-new",
-      expect.objectContaining({
-        modelIntent: "quality-optimized",
-        modelOverride: "claude-opus-4-6",
-      }),
+    expect(conversation.runAgentLoop).toHaveBeenCalledWith(
+      expect.any(String),
+      "msg-1",
+      expect.any(Function),
+      expect.objectContaining({ callSite: "analyzeConversation" }),
     );
   });
 
-  test("auto: does not pass modelOverride/modelIntent keys when config leaves them unset", async () => {
+  test("does not thread modelIntent/modelOverride into getOrCreateConversation", async () => {
+    // Per-call model selection now happens via the call-site resolver against
+    // `llm.callSites.analyzeConversation`, not via legacy modelIntent/
+    // modelOverride keys on the conversation create options.
     const conversation = makeConversation();
     const deps = makeDeps(conversation);
 
     await analyzeConversation("conv-1", deps, { trigger: "auto" });
 
-    const [, passedOpts] = (
-      deps.getOrCreateConversation.mock.calls as unknown as Array<
-        [string, Record<string, unknown>]
-      >
-    )[0] ?? ["", {}];
-    expect(passedOpts).toBeDefined();
-    expect("modelIntent" in (passedOpts ?? {})).toBe(false);
-    expect("modelOverride" in (passedOpts ?? {})).toBe(false);
+    const calls = deps.getOrCreateConversation.mock
+      .calls as unknown as Array<[string, Record<string, unknown> | undefined]>;
+    expect(calls.length).toBe(1);
+    const passedOpts = calls[0]?.[1];
+    if (passedOpts !== undefined) {
+      expect("modelIntent" in passedOpts).toBe(false);
+      expect("modelOverride" in passedOpts).toBe(false);
+    }
   });
 });

assistant/src/runtime/services/analyze-conversation.ts

@@ -8,15 +8,18 @@
  * Two triggers are supported:
  *   - **manual**: user-initiated analysis. Creates a fresh conversation each
  *     invocation, runs with `trustClass: "unknown"`, and strips the tool
- *     surface. Byte-for-byte unchanged from the original route logic.
+ *     surface.
  *   - **auto**: called by the auto-analyze job when a source conversation
  *     reaches a natural pause. Reuses a rolling analysis conversation per
  *     parent (creating one if none exists), runs with `trustClass:
  *     "guardian"`, and keeps the full tool surface so the analysis agent can
- *     write memory and skills directly. Reads optional model overrides from
- *     `analysis.modelIntent` / `analysis.modelOverride` config.
+ *     write memory and skills directly.
+ *
+ * Both triggers route the agent loop through `callSite: 'analyzeConversation'`
+ * so per-call provider/model selection flows through `resolveCallSiteConfig`
+ * against `llm.callSites.analyzeConversation` (populated by the unify-llm
+ * workspace migration from the legacy `analysis.modelIntent`/`modelOverride`).
  */
-import { getConfig } from "../../config/loader.js";
 import type { ServerMessage } from "../../daemon/message-protocol.js";
 import {
   AUTO_ANALYSIS_GROUP_ID,
@@ -31,7 +34,6 @@ import {
   getMessages,
 } from "../../memory/conversation-crud.js";
 import { resolveConversationId } from "../../memory/conversation-key-store.js";
-import type { ModelIntent } from "../../providers/types.js";
 import { getLogger } from "../../util/logger.js";
 import { buildAssistantEvent } from "../assistant-event.js";
 import { DAEMON_INTERNAL_ASSISTANT_ID } from "../assistant-scope.js";
@@ -173,8 +175,6 @@ export async function analyzeConversation(
   let prompt: string;
   let trustClass: "unknown" | "guardian";
   let stripTools: boolean;
-  let modelIntent: ModelIntent | undefined;
-  let modelOverride: string | undefined;
 
   if (opts.trigger === "manual") {
     const newConv = createConversation({
@@ -205,10 +205,6 @@ export async function analyzeConversation(
     prompt = buildAutoAnalysisPrompt(transcript);
     trustClass = "guardian";
     stripTools = false;
-
-    const analysisConfig = getConfig().analysis;
-    modelIntent = analysisConfig.modelIntent;
-    modelOverride = analysisConfig.modelOverride;
   }
 
   // h. Load the conversation into memory with the appropriate trust
@@ -219,13 +215,14 @@ export async function analyzeConversation(
   // Hoisted ahead of message persistence so the auto branch can detect a
   // still-running prior agent loop on the rolling conversation and bail out
   // before mutating any state. See concurrency guard below.
+  //
+  // Per-call model selection is no longer threaded through `getOrCreateConversation`;
+  // the runAgentLoop call below passes `callSite: 'analyzeConversation'` so the
+  // unified call-site resolver picks up provider/model from
+  // `llm.callSites.analyzeConversation`.
   const analysisConversation =
     await deps.sendMessageDeps.getOrCreateConversation(
       analysisConversationId,
-      {
-        ...(modelIntent !== undefined ? { modelIntent } : {}),
-        ...(modelOverride !== undefined ? { modelOverride } : {}),
-      },
     );
 
   // h.1. Concurrency guard (auto trigger only). The rolling analysis
@@ -298,11 +295,14 @@ export async function analyzeConversation(
   analysisConversation.abortController = new AbortController();
   analysisConversation.currentRequestId = crypto.randomUUID();
 
-  // l. Fire-and-forget the agent loop
+  // l. Fire-and-forget the agent loop. `callSite: 'analyzeConversation'`
+  // routes the per-call provider config through `resolveCallSiteConfig`
+  // against `llm.callSites.analyzeConversation`.
   analysisConversation
     .runAgentLoop(prompt, messageId, onEvent, {
       isInteractive: false,
       isUserMessage: true,
+      callSite: "analyzeConversation",
     })
     .catch((err) => {
       log.error(